Can you run this with DEBUG logs and paste in the master logs? Without a detailed description of what the master is doing, hard to diagnose.
Thanks! -ryan On Wed, Oct 14, 2009 at 3:32 PM, Yannis Pavlidis <ypavli...@oneriot.com> wrote: > Hi all, > > We are in the process of evaluating hadoop/hbase (using hadoop-0.20.1 and > hbase-0.20.0) for some of our needs and I wanted to ask a question related > with behavior I am seeing during our evaluation. > > I have a small cluster setup containing for your simplicity: server1, > server2, server3. Currently server1 runs the namenode for hadoop and server1, > server2 and server3 run the datanodes. > > In addition, server1 is the master and server1, server2 and server3 run > regionservers. > > During my testing I see in the master logs that ROOT table has been assigned > to server2 and META table has been assigned to server3 (I also have hbase > managing zookeeper) > > I run some of my tests and everything is working correctly (I am able to > put/get/delete in tables). > > Then I kill (simply kill <pid>) the regionserver on server1 (which holds > neither ROOT or META regions) and everything is working properly (run the > same suite). > > Then I kill (simply kill <pid>) the regionserver that has ROOT assigned to it > and it appears that the master never re-assigns the ROOT to another > regionserver (there is still one running which also has the META region). > > I can see the clients attempting to connect to server2 which is the > regionserver that had the ROOT region (which is not running any more) and > also I can see from web ui when I run the zkdump that the Region server > holding ROOT: is still server2. > > This seems to be very easily reproducible by following the above steps (I > have seen it getting reproduced with different server assignments of the META > and ROOT tables). > > I also tried using hbase-0.20-1 but I still seem to be getting the same > results. > > Any help would be highly appreciated. > > Thanks in advance, > > Yannis Pavlidis. >
